MEDINFO 2004 M. Fieschi et al. (Eds) Amsterdam: IOS Press © 2004 IMIA. All rights reserved
Morpho-Functional Visualization of Dynamic MR-Mammography K-H Englmeiera, G Hellwiga,b, J Griebelb, S Delormec, M Sieberta, G Brixb a
GSF – National Research Center for Environment and Health, Institute of Medical Informatics Ingolstädter Landstrasse 1, D – 85764 Neuherberg, Germany b BfS – Federal Office for Radiation Protection, Institute of Radiation Hygiene Ingolstädter Landstrasse 1, D – 85764 Oberschleissheim, Germany c DKFZ – German Cancer Research Center, Division of Oncological Diagnostics and Therapy Im Neuenheimer Feld 280, D – 69120 Heidelberg, Germany K-H Englmeier, G Hellwig, J Griebel, S Delorme, M Siebert, G Brix
breast imaging. Due to the realization that X-ray mammography – even if done as high-quality examination – is not sufficiently sensitive or specific as a stand-alone imaging technique [1], in recent years the use of several complementary technologies has increased significantly, in particular in the imaging assessment of younger women and for problem solving. First of all, ultrasound and dynamic contrast-enhanced magnetic resonance imaging (dMRI) have to be mentioned here. This paper takes a close look at breast MRI using modern approaches of image processing and visualization.
Abstract In view of an increasing use of breast MRI supplementing X-ray mammography, the purpose of this study was the development of a method for fast and efficient analysis of dynamic MR image series of the female breast. The image data sets were acquired with a saturation-recovery-turbo-FLASH sequence facilitating the detection of the kinetics of the contrast agent concentration in the whole breast with a high temporal and spatial resolution. In addition, a morphological 3D-FLASH data set was acquired. The dynamic image data sets were analyzed by tracer kinetic modeling in order to describe the physiological processes underlying the contrast enhancement in mathematical terms and thus enable the estimation of functional tissue specific parameters, reflecting the status of microcirculation. To display morphological and functional tissue information simultaneously, a multidimensional real-time visualization system (using 3D-texture mapping) was developed, which enables a practical and intuitive human-computer interface in virtual reality. The spatially differentiated representation of the computed functional tissue parameters superimposed on the anatomical information offers several possibilities: improved discernibility of contrast enhancement, inspection of the data volume in 3D-space and localization of lesions in space and thus fast and more natural recognition of topological coherencies. In a feasibility study, it could be demonstrated that multidimensional visualization of contrast enhancement in virtual reality is practical. Especially, detection and localization of multiple breast lesions may be an important application.
It has been noted that different pathologies (e.g., benign and malignant tumors) tend to exhibit different temporal patterns of contrast enhancement on dMRI of the breast, and the use of this observation in the assessment of tumors has been studied extensively over the past decade. (Compare e.g. [2, 3] for comprehensive references on this subject.) In general, malignant lesions show a stronger as well as earlier and more rapid enhancement than benign ones, but unfortunately most dMRI studies have shown varying degrees of overlap between enhancement curves derived from benign and malignant breast tumors. Since the kinetics of the sequence of events abbreviated as “wash-in – washout” mainly depends on physiological tissue parameters such as capillary permeability and surface area, plasma flow, extent of distribution volume in intravascular and interstitial space as well as the whole tissue volume in the region considered, the determination of a significant set of tissue specific parameters might help to overcome the above-mentioned issue. In principle, tracer kinetic modeling of contrast agent transport in tissues allows an estimation of functional tissue paameters in a non-invasive way and thus will play an increasingly important role for diagnostic imaging and follow-up studies. However, based on the over mapping patterns of tumor contrast enhancement, it is likely that there also will be some overlap between the microcirculation parameters of benign and malignant tumors. The extent of this overlap will determine the ultimate value of tracer kinetic modeling (compare [4]). From a practical point of view, the use of dynamic MR imaging is limited in the clinical settings due to the vast number of slices displaying either anatomical or functional information. The evaluation of this huge amount of information requires an enormous amount of time by the radiologist. Novel
Keywords Dynamic MR-mammography, tracer kinetic modeling, vitual reality.
Introduction World-wide, breast cancer is one of the most common malignancies among women; and for unknown reasons, its incidence still continues to increase each year. Up to now, early detection and treatment prior to metastasis is the only method of reducing mortality. Besides early detection of malignancy, its differentiation from other breast disease is a further major goal of diagnostic
838
K-H. Englmeier et al.
though information on displacement vectors in adjacent slices is taken into account.
approaches are highly warranted. Due to these issues, we present a multidimensional visualization system, which enables a practical and intuitive human-computer interface in virtual reality to display simultaneously morphological and functional tissue information of the female breast. To this end, several steps have to be considered: •
Measurement of the signal variation during and after (intravenous) application of a paramagnetic contrast agent and transformation of these observed signal-time courses to concentration-time courses.
•
Analysis of the concentration-time curves and specially estimation of functional tissue parameters by pharmacokinetic modeling as result of a mathematical approach describing the trans- and intercapillary passage of tracer particles.
•
High-quality volume visualization in real time using 3Dtexture mapping, embedded in virtual reality.
•
Combining pharmacokinetic breast MRI with conventional volume visualization in real time – possibly by adapting the representation of both the MR signal and the relevant set of model parameters as well as the size of the texture element to each other.
Pharmacokinetic Modeling For the analysis of the measured set of dynamic SRTF breast image series a pharmacokinetic two-compartment model is used. In this section a brief review of the model shown in Fig. 1 and originally proposed for the analysis of the signal enhancement of CNS lesions is presented, while the reader is referred to [6, 7, 8] for a detailed discussion of basic assumptions and limitations. For modeling the contrast agent application [constant-rate (Kin) infusion of duration τ ] and the ensuing kinetics of tracer concentration within the relatively short time range of dynamic MR measurements, a central blood-pool compartment with a rectangle impulse as input function and first-order kinetics for elimination may be assumed. Aside from the central (blood-pool) compartment (#1) the model consists of a peripheral compartment (#2) representing the combined intravasal and interstitial space of the lesion or, more generally, of the tissue under discussion. Both compartments are connected by linear exchange processes in both directions. Transport parameters as illustrated in Fig. 1 describe the velocity of the tracer transfer between the two compartments (k12 , k21) and the elimination (kel) of tracer particles from the central compartment.
Materials and Methods Data Acquisition MR imaging was performed with a conventional 1.5-Tesla whole-body MR system (MAGNETOM SP 4000; Siemens, Erlangen, Germany) using the circular-polarized body coil for RF transmission and a double-breast coil for RF detection. The patients lay prone on the breast coil with the arms extended above the head. In the first step, native 3D FLASH 32 transaxial images were acquired. On the basis of these static images, 15 transaxial sections were defined, from which 32 dynamic image data sets were measured over a period of 12 min (repetition every 23 s) using the dynamic Saturation-Recovery-Turbo-FLASH (SRTF) imaging sequence that was developed by Hoffmann and co-workers [5]. In contrast to conventional 3D FLASH sequences frequently employed in dynamic MR mammography, the specially optimized SRTF sequence has the advantage that the acquisition period for each image is known with an uncertainty of less than 1 sec. A total dose of 0.1 mmol Gd-DTPA (MAGNEVIST; Schering AG, Berlin, Germany) per kg of body weight was intravenously administered at a constant rate over 1 min using a variable speed infusion pump. Injection was initiated simultaneously with the beginning of the fifth measurement cycle. Finally, static 3D FLASH images were acquired using the same parameters as for the precontrast measurement
Figure 1 - Open two-compartment model with constant-rate infusion. Elimination occurs only from the central compartment. Morphofunctional volume-oriented 3D-visualization In order to present only morphological 3D-data, common 1component 3D-textures are sufficient: To every voxel of the data volume an intensity (gray) value is assigned and stored in the corresponding texel. For the volume-oriented visualization [6,7] of both the MR signal, i.e. the conventional gray value volume, and the results of the pharmacokinetic analysis, i.e. the relevant set of model parameters, the method has to be generalized and an adaptation of these quantities to the size of the texture element might be necessary. To this end, the tupel of bits representing the texture elements can be divided into two disjunct parts, one part representing the scaled gray values, the other the information of the pairs k21 , A) according to the cells in Fig. 2c. For every texel, this information is transferred into a color value via lookup table. More precisely, a 4-component (RGB) 3D-texture is used: the first three components belonging to the standard color vectors (red, green, blue) and the last one to the alpha-channel modeling transparency. This translation is done by the graphics subsystem without any performance penalty. As a result, a superposition of
Due to patient motion and changes in the shape of the breasts during the acquisition of the dMRI series, within the framework of pre-processing a coregistration of the data volumes was performed by means of an elastic matching algorithm before further image analysis in order to obtain high accuracy in the location of lesions. Up to now, the matching is actually slice-oriented, even
839
K-H. Englmeier et al.
Results:
morphological and functional information is created. This 3D visualization method is integrated in our immersive and intuitive virtual reality system which is based on volume and surface rendering, stereo projection and gesture analysis. As hardware components a head mounted display, a binocular omni orientation system, data gloves and 3d interaction devices are used.
Simultaneous display of morphological and functional tissue information in VR was performed in 6 patients in order to assess feasibility and preliminary information about the potential clinical impact of the development. Currently, the evaluation of a study with higher case rates is ongoing. Representative results for benign and malignant breast tumors are shown in Figs. 2 and 3. The spatially differentiated representation of the computed functional tissue parameters superimposed on the anatomical (morphological) information offers several possibilities. By comparison of conventional gray value volume visualization and morphofunctional 3D-visualization as illustrated in Fig. 2 (a,c), the better discernibility of the contrast enhancement in the left breast when including functional information is easy to recognize. All the manipulation of the data volume can be done by gesture-driven interaction which is performed by the use of a data glove and the application of agesture analysis system. Figure 3 shows different views of the breasts of a patient with multiple fibroadenoma. In such a case the advantage of the morphofunctional 3D-visualization in real time is evident: A comprehensive inspection of the data volume in the “common” three-dimensional space using the features of rotation and transparency variation becomes possible. Lesions can be displayed and distinguished quite pleased.
(a)
In summary, a 3D representation of the functional tissue parameters as well as a quite exact localization of lesions in space is possible. Topological coherencies can be recognized in a much faster and more natural way, which may be an advantage specially in the case of multiple breast lesions. Finally, 3D-visualization offers a fast and efficient overview in compressed form before studying special slices with special regions of interest.
(b )
(c) Figure 2 - Morphofunctional visualization in our VR environment. a / b - Comparison of gray value volume visualization and morphofunctional 3D-visualization. c: Color coding of the lesion specific parameters A and k21 , chosen as follows: Processes with a more rapid and stronger contrast enhancement are brightly colored, while slow and minimal enhancing issues appear in dark colors
Figure 3 - Four different aspects of the morphofunctional volume-oriented 3D-visualization (multiple fibroadenoma). Line of vision and transparency can be varied in real time
Discussion, Conclusion The approach carried out in this paper combines volume-oriented 3D-visualization using 3D-texture mapping with the interaction techniques of virtual reality in order to present the results of the pharmacokinetic analysis. In order to prevent misunderstandings concerning the aim and object of the development car-
840
K-H. Englmeier et al.
[6] Haubner M, Krapichler C, Lösch A, Englmeier K-H, van Eimeren W. Virtual Reality in Medicine – Computer Graphics and Interaction Techniques. IEEE Transactions on Information Technology in Biomedicine 1997; 1(1): 61 – 72.
ried out, the following concluding remarks, pointing out possibilities and limitations, should still be emphasized. Like with all visualization tools, the clinician must be aware that the viewed images are created by eliminating contents from the source images that possibly obstructs the display of the lesion in question. Basically, the processed data cannot contain more information than the source data. Therefore, viewing dMRI studies in VR will not obviate careful reading of the source images until it is prospectively proven that the processed data alone are sufficient. Particularly, for discriminating between benign and malignant lesion, the reader will still need to analyze its morphology (e.g., shape, architectural distortion, homogeneity) in detail, and even refer to additional findings other than MRI (clinical findings, X-ray mammography, ultrasound, biopsy). Theoretically, no diagnosis should be possible reading dMRI studies in VR, which could not be obtained from the parameter maps alone, of course combined with the static pre- and postcontrast studies. However, computer aids in visualizing contrast enhancement in VR may prove beneficial for the following reasons: 1. Obtaining an overview is simply faster and easier, particularly concerning presence or absence of disease, or gross asymmetries. Furthermore, 3D-visualization in VR may focus the attention of the radiologist to regions, which should be examined in detail by using the source images – both anatomical and functional. 2. The spatial arrangement of pathological disease is often difficult to understand from reading single images alone. This is of particular importance when trying to discriminate whether a given, complex lesion is either a contiguous one, or rather the result of two neighboring, multifocal tumors. 3. It is notoriously hard to communicate imaging findings to clinicians. Even for radiologists experienced in cross-sectional imaging techniques, it is sometimes difficult to relocate a lesion seen on MR studies, when confronted with the patient herself. A common “workaround” is to try and reproduce the lesion using ultrasound. Techniques for visualizing a given lesion in loco three-dimensionally will definitely be helpful, although still differences in positioning (prone in the MR scanner, supine for clinical examination or surgery) will cause some problems and – should the situation arise – preoperative localization will be necessary in future as well.
[7] Krapichler C, Haubner M, Engelbrecht R, Englmeier K-H. VR interaction techniques for medical imaging applications. Computer Methods and Programs in Biomedicine 1998; 56: 65 – 74. Adress for correspondence Karl-Hans Englmeier, GSF – National Research Center for Environment and Health, Institute of Medical Informatics, Ingolstädter Landstrasse 1, D – 85764 Neuherberg, Germany,
[email protected]
References [1] Jellins J. Challenges in Quality Assurance. In: Madjar H, Teubner J, Hackelöer B-J, editors. Breast Ultrasound Update. Basel – Freiburg: Karger, 1994; 13 – 25. [2] Heywang-Köbrunner SH, Beck R. Contrast-Enhanced MRI of the Breast. Berlin – New York: Springer, 1996. [3] Heywang-Köbrunner SH, Dershaw DD, Schreer I. Diagnostic Breast Imaging. Stuttgart – New York: Thieme, 2001. [4] Henderson E. Measurement of Blood Flow, Blood Volume and Capillary Permeability in Breast Tumours using Contrast-Enhanced Magnetic Resonance Imaging. University of Western Ontario: PhD thesis, 1999. [5] Hoffmann U, Brix G, Knopp MV, Heß T, Lorenz WJ. Pharmacokinetic Mapping of the Breast: A New Method of Dynamic MR Mammography. Magnetic Resonance in Medicine 1995; 33: 506 – 514.
841
